High frame rate contrast enhanced ultrasound imaging for slow flow: influence of ultrasound pressure and flow rate on bubble disruption and image persistence

Contrast-enhanced ultrasound (CEUS) utilising microbubbles shows great potential for visualising lymphatic vessels and identifying sentinel lymph nodes (SLNs) which are valuable for axillary staging in breast cancer patients. However, current CEUS imaging techniques have limitations that affect the accurate visualisation and tracking of lymphatic vessels and SLN. (i) Tissue artefacts and bubble disruption can reduce the image contrast. (ii) Limited spatial and temporal resolution diminishes the amount of information that can be captured by CEUS. (iii) The slow lymph flow makes Doppler-based approaches less effective. This work evaluates on a lymphatic vessel phantom the use of high frame rate (HFR) CEUS for the detection of lymphatic vessels where flow is slow. Specifically, the work particularly investigates the impact of key factors in lymphatic imaging, including ultrasound pressure and flow velocity as well as probe motion during vessel tracking, on bubble disruption and image contrast. Experiments were also conducted to apply HFR CEUS imaging on vasculature in a rabbit popliteal lymph node (LN). Our results show that (i) HFR imaging and singular value decomposition (SVD) filtering can significantly reduce tissue artefacts in the phantom at high clinical frequencies; (ii) the slow flow rate within the phantom makes image contrast and signal persistence more susceptible to changes in ultrasound amplitude or mechanical index (MI), and an MI value can be chosen to reach a compromise between images contrast and bubble disruption under slow flow condition; (iii) probe motion significantly decreases image contrast of the vessel, which can be improved by applying motion correction before SVD filtering; (iv) the optical observation of the impact of ultrasound pressure on HFR CEUS further confirms the importance of optimising ultrasound amplitude and (v) vessels inside rabbit LN with blood flow less than 3 mm/s are clearly visualised.


Introduction
contrast agents to allow better visualisation of the vasculature and flow. When microbubbles are 23 exposed to ultrasound field, they produce backscattered signals containing fundamental and harmonic components, which greatly increase the ultrasound echoes from within the vessels. 1 Microbubbles have been widely used in clinical diagnosis and are safe with little reported side 2 effect or complications. CEUS allows the preoperative identification of SLN and when 3 combined with image guided needle biopsy, serves as a promising axillary nodal staging 4 method.

5
There are several studies on the use of CEUS for the lymphatic system. The technique 6 was first described by Goldberg et al. in several swine melanoma model (Goldberg et al. 2004). 7 Sever et al. (Sever et al. 2011) utilised CEUS for the visualisation of lymphatic vessels and 8 SLNs in patients with diagnostic breast cancer. After the periareolar intradermal injection of 9 microbubbles, they successfully monitored the complete process of lymph draining which starts 10 at the afferent lymphatic vessel and flows through the lymph node and finally exits through the 11 efferent vessel. In their most recent study (Cox et al. 2013), CEUS was used to guide 12 percutaneously SLN biopsy after preoperative detection of SLN in 295 breast cancer patients. 13 With the help of CEUS imaging, a secondary surgical procedure was avoided in 29 patients.
14 Although the feasibility of SLN identification by tracking microbubbles transporting 15 through lymphatic vessels was confirmed by these encouraging results, tissue artefacts and 16 bubble disruption can reduce the contrast of lymphatic vessels, compromising the image quality. 17 The failure rate of SLN detection ranges between 4% (Cox et al. 2013) and 11% (Sever et al. 18 2011) even with experienced operators. Different CEUS imaging techniques such as pulse 19 inversion (PI) (Simpson et al. 1999) and amplitude modulation (Brock-Fisher et al. 1996) are 20 developed to distinguish microbubble and tissue signals by detecting nonlinear echoes arising 21 from microbubble oscillations. However, nonlinear echoes can also be produced by nonlinear 22 propagation of ultrasound in tissue or bubble cloud (Tang et al. 2010). This can misrepresent 23 tissue as microbubbles in the final image, forming nonlinear tissue artefacts. These artefacts 24 reduce image contrast, particularly at high clinical ultrasound frequencies where the sensitivity disruption and image contrast, particularly in the context of lymphatic flow, have not been 1 investigated.

2
In this study, we apply SVD filter on HFR CEUS for the visualisation of slow flow on a 3 lymphatic mimicking phantom. In-vitro experiments on lymphatic mimicking phantom are 4 designed to understand the impacts of lymph flow velocities, ultrasound pressures and probe 5 motion on HFR CEUS image contrast. Our study indicates the trade-off between minimizing 6 microbubble disruption and achieving required contrast for detecting microbubble from tissue 7 signals, and shows the need for optimising ultrasound pressure for different flow velocities. 8 HFR and optical microscope were used to observe the impact of the MI of HFR ultrasound on 9 bubble disruption. Preliminary studies were conducted on imaging a rabbit lymph node 10 microvasculature.

12
In vitro experiment set-ups and ultrasound parameters 13 A lymph vessel mimicking phantom (Fig. 2a) was made consisting of a 300 µm tube 14 (Platinum Cured Silicone Tube, Harvard Apparatus UK, Cambridge, United Kingdom) inside a 15 tissue-mimicking phantom (Madsen et al. 1998), which was made of agar containing glass beads 16 (45-90 µm in diameter, Potters Ballotini Ltd, Barnsley, South Yorkshire, United Kingdom) as 17 scatterers. The diameter of the tube was chosen according to the average size of the afferent 18 lymphatic vessels in human (Pan et al. 2010). The average depth of tube was set to 1.6 cm from 19 the top surface, which mimicked the average depth between the axillary lymphatic vessel and 20 the skin (Bentel et al. 2000). Flowing microbubbles of concentration ~2.5x 10 7 MB/mL were 21 administered using a syringe pump. The microbubbles used for all the experiments were inhouse lipid-shell microbubbles filled with decafluorobutane which were manufactured 23 according to the formula described in (Lin et al. 2017).
HFR plane wave imaging data were acquired using a research platform (Verasonics 1 Vantage 128, Verasonics, Kirkland, USA), a linear probe L12-3v (192-element linear array 2 transducer, Verasonics, Kirkland, USA) and a 5-angle compounding plane wave imaging 3 sequences. The transmitted pulses were centred at 4 MHz with a pulse repetition frequency of 4 5000 Hz for all the ultrasound acquisitions. The acquired radio frequency (RF) data were 5 beamformed to yield 1000 frames of compounded B-mode images per second which were then 6 processed by SVD filter. All in vitro experiments in this paper are repeated three times unless 7 otherwise stated.

8
After SVD, microbubble and tissue signals can be represented by different sets of 9 singular vectors with independent spatiotemporal coherence. Among these vectors, large 10 singular values correspond to tissue signals with high temporal and spatial coherence, while 11 lower components are associated with blood signals with relatively low spatiotemporal 12 coherence. Based on this assumption, distinction between clutter and blood/microbubble signals 13 can be achieved by setting threshold on the singular values. For all the B-mode images in this 14 study, components corresponding to higher singular values, which represent tissue signals, were 15 filtered out. The SVD cut-off thresholds were chosen so as to achieve the maximum contrast-16 to-tissue ration (CTR) (Song et al. 2017). 17 In order to compare the performance of SVD in combination with HFR imaging and 18 standard CEUS technique -PI in suppressing tissue artefacts, a two-tube set-up ( Fig. 2b) was 19 made consisting of a 200 µm cellulose tube (Cuprophan,RC55 8/200 Membrana GmbH,20 Germany) filled with non-flowing microbubbles, and another tube filled with water as control.

21
Plane wave B-mode images were obtained with the same ultrasound settings and were 22 subsequently processed by SVD filter. Plane wave PI images were acquired with the same For the images acquired with probe motion, two different processing methodologies 1 were applied and investigated. One of the methods utilised SVD filter which was the same as 2 that used for images from the static probe. The other method corrected for the motion induced 3 by the moving probe before application of SVD filter. The motion correction was performed 4 based on a rigid image registration algorithm (Harput et al. 2017). The motion estimation was 5 performed on the original B-mode image after normalisation and log-compression of the 6 complex magnitude of beamformed data. Then the transformation matrix was applied to the real 7 and imaginary part of the corresponding beamformed data separately. SVD filter was 8 subsequently used to process the motion-corrected complex data.  angle, a single-cycle plane wave pulse was transmitted at 18 MHz and a mechanical index of 0.11. The acquired radio frequency (RF) data were beamformed to yield 800 frames of 1 compounded B-mode images which were then processed by SVD filter. The filtered data were 2 temporally correlated to generate high contrast vascular images. Furthermore, estimation of 3 frequency shift generated the direction and velocity of blood flow inside the imaging node 4 which were shown on directional image. where N is the total frame number. The ability for retaining contrast signals over time regardless 19 of initial amplitude can be captured by INI due to the normalisation process.

20
The region of interest (ROI) for microbubbles was chosen according to the shape of the 21 tube in the lymphatic vessel phantom. The ROI for background tissue was set to be 3 mm above 22 that of ROI for microbubbles. In the experiment for investigating the effect of flow velocities and ultrasound pressure, both ROIs were fixed throughout the sequence as there was no motion 1 among all the acquired frames. In the experiment for the effect of probe motion, the same fixed 2 ROIs were used in the case of images after motion correction associated with SVD filter. But

18
In vitro ultrasound experiment: comparison between PI and SVD filter 19 SVD filter and PI techniques were qualitatively compared in terms of tissue artefacts. In We then evaluate the effect of applying motion correction before SVD filter. Fig. 10b 1 reported the CTR before and after the application of motion correction algorithm. For the lower 2 flow velocities of 4 mm/s and 10 mm/s, motion correction results in higher CTR.   In terms of ultrasound pressure, there is always a trade-off between microbubble 1 disruption and required contrast for detecting microbubble from tissue signals. Disruption of 2 microbubbles by different pressures of plane wave imaging has been acoustically studied by 3 Olivier et al (Couture et al. 2012). Compared to conventional focus imaging, plane-wave 4 imaging achieved better contrast at a specific disruption ratio. However, the relative effects of 5 flow rate and ultrasound pressure have not been evaluated. It appears that the effect of ultrasound 6 pressure differs under various flow velocities.

7
Enhanced contrast with higher MI can be found at 50 ms with both velocities of 4 mm/s 8 and 10 mm/s. However, towards the later stage of the acquisition at 550 ms, we observed better 9 contrast using moderate MI (0.11) at 4 mm/s. At 10 mm/s, higher MI still gives better contrast.

10
This explains the different results of CTR according to MI under distinct flow velocities. In Fig.   11 8a, at 4 mm/s, the highest value of CTR is given by the moderate MI and the lowest value is 12 acquired by the lowest MI. In other cases of higher flow, higher CTRs are achieved by higher 13 MI. This is because increase in ultrasound pressure introduces varied microbubble disruption 14 behaviour, especially with large amount of microbubbles at the beginning of the acquisition.

15
These microbubble changes are able to produce dissimilarity in temporal characteristics and 16 spatial distribution (i.e. spatiotemporal coherence) between microbubble and tissue signals, 17 resulting in improved efficiency of STA. A higher MI gives much higher contrast signals at the 18 beginning of the acquisition due to the violent and rapid changes of microbubbles. It also tends 19 to destroy the bubbles faster, causing contrast signals to decrease sharply and they cannot be 20 retained over a long time as shown in Fig. 6f faster replenishment of microbubbles compensates for the decay of signals from microbubbles due to disruption. Thus, signals from microbubbles can last longer until the end of data 1 acquisition. This explains both the qualitative results in Fig. 7 and results of CTR under flow 2 velocities from 10 mm/s to 80 mm/s.

3
Based on these qualitative and quantitative evaluation, it can be concluded that the 4 image contrast is dependent on the combined effects of ultrasound pressure and flow velocity.

5
In addition, the ability to retain contrast signals over time (i.e. persistence of contrast) is another there is significant reduction of CTR for the moving probe for all the flow velocities as shown in Fig. 9 and Fig. 10a. This is because moving probe/tissue may share comparable 23 spatiotemporal characteristics with flowing microbubbles. Hence, the suppression of clutter 24 signal by STA is less efficient with probe motion especially for the slow flow velocities (< 10 mm/s). For this case, the efficacy of clutter filtering is improved with the help of motion 1 correction, where tissue motion resulting from moving probe is estimated and corrected first to 2 reduce the overlap in spatial and temporal domain between clutter and microbubble signals. This 3 allows subsequent SVD filter to be more efficient in discriminating between echoes from tissue 4 and microbubbles. The fast-flows on the other hand, even with probe/tissue motion, have 5 adequate difference from tissue signals both in space and time for SVD filter to work.
6 Accordingly, improvement is not found with the help of motion correction at higher flow rates.

7
It should be noted that the improvement of SVD filter in terms of filtering clutter signals 8 under probe motion, in comparison with classic nonlinear sequence and high pass filters, has 9 been shown in CEUS imaging (Desailly et al. 2017). In that study, much higher CTR was 10 obtained at the same probe speed of 2 mm/s. That is likely due to the different application focus 11 in that study the diameter of the wall-less vessel in that tissue phantom was 5 mm which was 12 much larger than the 300 µm vessel in this lymphatic mimicking phantom. The contrast signal  Postema et al. 2002;Postema et al. 2004b;Postema et al. 2004a;Zheng et al. 2007).

19
In this paper, we employ similar optical method to study the signal persistence of microbubbles 20 under slow flow conditions when exposed to ultrafast ultrasound plane wave imaging with a studies (Baranger et al. 2018;Song et al. 2017).
The effects of other scanning parameters such as frame rate and frequency have not been 1 included in this study and are worth investigating in future work. For instance, the frame rate 2 gives a trade-off between the ability of capturing the slow lymph fluid and increased 3 microbubble destruction. Furthermore, SVD filter can be further optimised in terms of 4 processing region and singular value threshold determination for in-vivo human imaging.

5
Instead of global SVD, block-wise filtering technique can be used for the complicated spatial 6 distribution of tissue and noise (Song et al. 2017). Adaptive methods for optimal thresholding 7 can also be used to compare the efficiency in clutter and noise suppression (Baranger et al. 2018;8 Song et al. 2017).

9
Due to the practical difficulties associated with injecting the microbubbles into lymphatic